Connect your intelligence

Weather intelligence.
In your AI workspace.

Bring degree-day forecasts, model revisions and historical context into the conversations where you work.

Read-only toolsGoogle sign-inModel Context Protocol

Connect with ChatGPT

Web · custom MCP connection
Supported client
MCP server URL
https://mcp.auraflowllc.com/mcp/

Use this endpoint—not the address of this guide. Transport: Streamable HTTP.

  1. Add a custom connection

    In ChatGPT on the web, open Plugins → + → Add custom MCP server. On the older interface, use Settings → Apps → Create.

    Your account or workspace may require administrator approval for custom connections.

  2. Enter the connection details

    Name it AuraFlow Degree Days and paste the server URL above. Choose OAuth; leave the client ID and client secret empty.

    If registration options appear, choose Client ID Metadata Document (CIMD).

    Authentication · OAuth
  3. Sign in securely

    Review the custom-server prompt, create/install the plugin, then choose Sign in with Google and approve read access. No shared API key or copied access token is needed.

  4. Start a conversation

    In a chat, type @, select AuraFlow Degree Days, and try one of the questions below. Your assistant will call the relevant read-only tools.

Make your first request

Start with a specific model and measure. Your assistant can help you explore from there.

Track the revisions

Show the latest GFS HDD changes and the dates driving the largest revisions.

degree_day_changes

Compare the daily outlook

Compare the latest GEFS daily HDD curve with prior cycles and the 10-year and 20-year normals.

daily_degree_day_curve

Follow the monthly picture

Compare GFS and AIFS HDD estimates for the current and next month. Identify any missing reference dates.

monthly_degree_day_evolution

Read the weather briefing

Get the latest WPC weather briefing. Show its issuance and valid dates, temperature outlook and model uncertainty, and tell me if it is stale.

get_recent_weather_briefing
If connection fails
  • Check that the server URL ends in /mcp/, not /mcp/setup, and that OAuth is selected. Opening the endpoint in a browser is not a connection test.
  • If custom connections are unavailable, check your ChatGPT account’s interface and workspace policy. See OpenAI’s custom MCP guide for requirements.
  • If sign-in or permission approval fails, reconnect using your Google account and contact the service operator if access is still denied. This guide is public; forecast tools require authentication.
  • Do not copy browser tokens or use shared operator credentials to connect another agent.
Data sources and interpretation

Forecasts and ERA5 10-year/20-year normals are real. Monthly elapsed dates use an IFS analysis/short-range reference, with temperature interpolated to hourly—not measured station observations. Missing dates stay gaps; no mock fallback is used.

Revision sigma values are configurable and uncalibrated. Recent model-cycle coverage can differ; the service retains up to 16 recent cycles per model. Historical archive browsing is unavailable. Tool responses carry source contributions and coverage.

Model identifiers: gfs, aifs-ens, gefs, ifs. Measures: hdd, cdd. Defaults: GFS, HDD, CONUS, population weighting, 65°F base and fixed UTC−05:00 valid days.

Account access and availability

Dashboard and AI-agent access are separate entitlements: trader for Grafana and mcp for agent connections. Signup currently includes both at no additional charge during the trial. AI-agent access is planned as a separate paid plan; a dashboard login alone will not grant it.

Raw JSON/text endpoints are private. Separate paid programmatic/archive access is planned; billing is not enforced in this trial. Available models and references depend on the connected deployment’s data.